Redistribution of Routing Update Traffic

The term redistribution is technically defined as the exchange of routing update packets from one routing protocol domain to another routing domain. All the Cisco-supported routing protocols support route redistribution. Route redistribution may be necessary in any of several scenarios where an organization may be implementing multiple protocols. One may be that you are using products from several vendors , including Cisco devices. You are using a common open standard protocol such as RIP or OSPF for the non-Cisco devices and perhaps IGRP and EIGRP for the Cisco routers. Or perhaps you are migrating to a new routing protocol altogether, yet you need to retain the older protocol for awhile to accommodate certain areas of your network that are unable to migrate at the same time. You may have to implement several redistribution boundaries until the migration has been completed. Some areas may simply never migrate to a newer IGP for a variety of reasons. Some business units might require RIP for host-based routing and want to keep it for their LAN networking. Also, if a department's filtering policy does not fit into the long-term network infrastructure, you might need to isolate that department from the other routing protocols on your internetwork.

In the context of this chapter, an autonomous system represents internetworks running different routing protocols ”IGPs or EGPs. Do not confuse this with the Autonomous System as an entity used with Border Gateway Protocol.

Selecting the Optimal Path

Because routing metrics and algorithms are inherently incompatible, the process of determining the best route when multiple routing protocols are in use is tantamount. There are actually two steps involved. First, if there is more than one source (static or dynamic) for any destination network, the router decides which of the two sources has the lowest administrative distance. Secondly, after the routing process is selected (usually a dynamic routing protocol), the route provided by that particular routing protocol with the best metric is selected.

Cisco utilizes two mechanisms for choosing the best path when one or more routing protocols are in play. The first is the administrative distance (AD) value. It is proper to re- emphasize why administrative distance is so important. The worst thing that can happen in a routing domain is a routing loop, which can cause data black holes and a nightmare for the network engineer. The easiest solution for preventing routing loops is to avert route feedback through route filtering. This is ineffective , however, because it can prevent less desirable, though useable, routes from being implemented in the event of a primary path failure. The next easy solution is to change the metric for all routes being redistributed. Although this is a good solution, it is subject to failure under various (remotely possible) circumstances.

The most elegant way to prevent routing loops for redistributed routes is to change the administrative distance of those redistributed routes to be worse (higher) than the administrative distance of the protocol that is offering the best path. For example, you can change the AD of routes (originally in IGRP) to 150, which is worse than the AD of RIP (120). Even though IGRP provides superior metric calculation, because these routes are being redistributed, the metric value is typically inaccurate and can lead to suboptimal path selection. By making one or more redistributed IGRP routes less desirable, you ensure that the more accurate (shorter) RIP routes get used. However, you also ensure that in the event of a primary path failure, the less desirable (but fully useable) backup path can be used. The administrative distance is an integer value between 1 and 255 that designates the trustworthiness of a certain route's source. The AD value is generated from the metric of a dynamic routing protocol or is manually overridden by the administrator. A router trusts, and then injects into the routing table, routes with lower AD values before routes with higher AD values. You can see the default AD values of routing protocols in Table 10.2.

Table 10.2. Default Administrative Distance Values of the Cisco-Supported Routing Protocols

Source of Route	Default Administrative Distance
Connected interface
Static route outbound on interface
Static route to next hop	1
EIGRP summary route	5
External BGP	20
Internal EIGRP	90
IGRP	100
OSPF	110
IS-IS	115
RIPv1, RIPv2	120
EGP	140
External EIGRP	170
Internal BGP	200
Unknown	255

The second mechanism for selecting the optimal path is the routing metric. This value measures the path ”usually hop count or cost ”between the source router and the destination network. Because routing protocols implement different metrics, you modify the metric when redistributing routes. For instance, when going from RIPv2 to OSPF, you need to translate the metric from an RIP hop count to an OSPF cost.

Implementing Successful Route Redistribution

Although we have covered a variety of routing protocols in great detail, you need to consider in depth the techniques for connecting multiple networks running different routing protocols. In today's dynamic environment, mergers, acquisitions, corporate politics, security, and multiple protocols often require that more than one routing protocol be used, at least for a short time. A common misconception is that if you simply connect two disparate networks together, the router automatically redistributes routes between them. Although redistribution occurs automatically between IGRP and EIGRP if the autonomous system numbers are identical, this is not the case for the other routing protocols covered in this book. Redistribution is a necessary task that the network engineer must accomplish when connecting systems that use multiple routing protocols.

Redistributed routes must have a seed value (default or manually configured). For routers with directly connected networks, metric costs are determined by the routing protocol activated on the interface connected to the network. For redistributed networks (and subnets), there is no "direct connection"; therefore, a metric value must be assigned (seed value) to redistributed routes, to allow propagation to provide appropriate (not necessarily shorter) route selection. A router that is directly connected to the destination network typically derives the seed metric . Because a redistributed route is not physically connected to an ASBR, however, and different routing protocols have different metric values, you may need to configure the seed metric manually when distributing RIP or IGRP by using the default-metric router configuration command.

After the seed metric is configured for a redistributed route, the metric is incremented in a normal manner (unless it is the Type 2 external route in OSPF) as the route moves through the autonomous system. To prevent route feedback (loops), it is preferable to set the seed metric value higher for redistributed routes than the one used for a redundant route that originated in the core routing AS. It is rare for a preferred path to be through an external AS; therefore, the seed metric should be less desirable than the longest path in the receiving AS.

It is important to realize, however, that redistribution raises the level of complexity considerably for managing an internetwork and should be implemented only after much planning and forethought. The number of routing loops can increase when you integrate redistribution on boundary routers. The goal of this section of the chapter is to learn how to ensure optimized routing while eliminating routing loops. You also have to consider that different routing protocols have different convergence times and the impact that this will have on updates and the smooth flow of data packets. The impact is suboptimal path selection or a routing loop. Finally, redistributed routes introduce suboptimal path selection because the seed metric value (either correctly or incorrectly configured) can make a redistributed (and less desirable) route look more attractive to the router logic. This potential for network complexity should lead you to the conclusion that route redistribution should be used only when absolutely necessary.

Even though all the Cisco-supported routing protocols support route redistribution, generally speaking, you can redistribute only protocols that support the same protocol software. For example, you can redistribute between OSPF and IP RIP because both protocols support TCP/IP. Attempting to exchange routes between IPX RIP and OSPF is a different story because OSPF does not support the IPX/SPX protocol stack necessary to run IPX RIP.

The most important principle to follow is to be familiar with your network design, traffic, and business model. This includes being cognizant of impending changes in the infrastructure such as mergers or acquisitions.

Because IGRP and EIGRP have such a similar metric, redistribution will be automatic (if AS numbers are the same) when migrating to EIGRP. This fact led some organizations to prematurely rush into an EIGRP migration when, for business purposes, OSPF may have been the better long-term choice.

Another good piece of advice is to avoid running multiple routing protocols in the same internetwork. It is imperative that you draw distinct boundaries between routing domains (networks) that are implementing different routing protocols.

As mentioned earlier in this chapter, one-way redistribution is less complex to administer and troubleshoot. If you decide to implement a complex two-way redistribution design, as shown in Figure 10.2, you need to implement policy mechanisms such as default routes for a one-way redistribution, route filtering, metric tweaking, and the modification of administrative distance to prevent routing loops.

Figure 10.2. A rather elaborate internetwork that requires planning to prevent routing loops and convergence issues.

Some specific circumstances dictate when to avoid route redistribution. For example, in redistributed situations in which both the core ( backbone ) and the edge routing protocols (the protocols from which you are redistributing) advertise paths to the same destination network, you are likely to apply methods for avoiding certain routes from being exchanged and propagated between domains, such as filtering network packets as they enter a router's inbound interface. However, access lists alone are not meant to be used as route update filters. Both distribute lists and route maps use standard or extended access lists to control route update traffic. You can also influence route selection by modifying the administrative distance values or using passive interfaces, default routes, or static routes.